Security researchers at Israeli cybersecurity firm Dream have documented what experts are calling the first fully autonomous AI cyberattack against a government — a four-day operation against Taiwanese government systems in early July, suspected to be linked to China.
What the Attackers Built
The attack framework wasn’t a piece of sophisticated state-developed malware. It was assembled from publicly available tools. The operators used open-source AI agent frameworks — including Hermes, developed by Nous Research, and OpenClaw — to build a multi-agent system capable of mapping networks, discovering vulnerabilities, attempting intrusions, and adapting tactics automatically when one approach hit a wall. At peak activity, up to eight agents ran in parallel, each assigned to a different target or attack technique, operating more like a coordinated digital strike team than a script running a pre-programmed sequence.
The campaign unfolded between July 1 and July 4, 2026. Across twelve documented “attack waves,” the framework mapped 21 connected government systems, cracked 85 employee accounts, and exfiltrated at least 2,564 personnel records. Researchers retrieved a 160-megabyte archive of 1,395 files from the attackers’ operational workspace — including seven single sign-on client secrets, six internal database credentials spanning three different database systems, and internal network IP ranges. After cracking 85 accounts, the agents tested each one against connected internal systems through SSO; 84 of the 85 compromised credentials successfully accessed an internal information system because the SSO bridge trusted existing sessions without requiring additional authentication or multi-factor verification.
What Made It Autonomous in a Meaningful Sense
The system didn’t just execute pre-written instructions faster than a human could. It conducted what researchers called “learning cycles” — autonomous sessions in which the agents searched vulnerability databases, GitHub repositories, and security research publications to identify specific techniques and CVEs applicable to the targeted infrastructure, then adjusted their approach based on what they found. When one attack path was blocked, the agents reranked all available paths by probability of success using a Bayesian prioritization model and automatically switched to the next best option. The operation then expanded beyond initial targets to scan government IT supply chain vendors, a nuclear safety agency, a government email system, and at least seven energy companies — all in parallel.
Attribution and the Public Evidence
Dream’s researchers do not formally attribute the attack to the Chinese government or any specific hacking group, but note that internal operational documentation was written in simplified Chinese, pointing to a Chinese-language operator. Taiwan’s Ministry of Digital Affairs later confirmed an AI-assisted attack on government agencies in July. A spokesperson for China’s Ministry of Foreign Affairs told CNN it was not familiar with the situation.
Kenny Huang, chairman of the Taiwan Network Information Center, told CNN he believes this is the first disclosed case of a fully automated attack against a government. “It spells out one thing loudly,” Dream wrote in a post summarizing the findings: “the cost of running a competent attack has collapsed, but the cost of defending against one has not.”
Why This Changes the Threat Calculation
Prior AI-assisted cyberattack cases involved AI helping human operators write code, craft phishing emails, or search vulnerability databases. This operation reportedly required no ongoing human intervention during execution — once deployed, the agent framework pursued its objectives, learned, adapted, and expanded scope on its own across four days. That shift from AI-assisted to AI-autonomous attacks is exactly the scenario CISA and allied agencies have been warning about, particularly as the same week produced a federal emergency patch order for a critical vulnerability in the Ray AI compute framework widely used across the AI industry. See our coverage of that CISA advisory for the defensive side of the same dynamic.
What to Watch Next
The Taiwan incident is likely to accelerate calls for international norms around autonomous offensive AI systems, a governance gap that both the EU and US regulatory frameworks have not yet directly addressed. Whether similar operations are already underway against other governments — and have simply not yet been discovered or disclosed — remains an open and unsettling question.
Sources: Financial Times via Hackread, CNN, The Register
Disclaimer: This content is meant to inform and should not be considered financial advice. The views expressed in this article may include the author’s personal opinions and do not represent Times Tabloid’s opinion. Readers are advised to conduct thorough research before making any investment decisions. Any action taken by the reader is strictly at their own risk. Times Tabloid is not responsible for any financial losses.
